[tannevaled]

What

The Known limitations section listed nothing that is still a limitation.
Measured at v0.11.0 on 2026-10-04:

the section said	measured
BEGIN { } — "does not parse"	parses
END { } — "does not parse"	parses
CRuby stdlib: 46 of 47	47 of 47
the one refusal: mkmf.rb	parses

mkmf.rb tripped on def self::log_open — the def self::name form — which
parses now, and that is what took the stdlib figure to 47 of 47.

The section was also dated "as of v0.4.0", seven minors back.

Why this is worse than a stale number

A section naming limitations that no longer exist sends a reader round a bug
that is fixed: they avoid BEGIN/END for nothing, or go looking for the
mkmf.rb refusal. A number that is merely out of date is read as approximate;
a named limitation is read as a fact.

The claims are now executable

• TestReadmeClaimsStillHold parses every construct the section names, so
  a future change that breaks one fails the suite rather than quietly
  contradicting the page.
• TestUnderPermissiveNotOver pins the design claim with its discriminating
  case: MRI refuses BEGIN inside a method body, and so must this. A test that
  only checked things we accept could not tell under- from over-permissive.
• TestCRubyStdlibTopLevelParses runs the sweep that produced the 47-of-47
  figure, and skips when no CRuby tree is present. CI has none, which is
  exactly why the README states that figure as a dated local measurement
  rather than as a gate — with the command to reproduce it.

Before writing "none"

I probed 23 further constructs against ruby -c — flip-flops, redo,
__END__/DATA, heredoc-with-method-call, _1, it, endless methods,
rightward assignment, one-line pattern match, hash shorthand, all three
anonymous forwards, refinements, ?a, imaginary/rational literals, special
globals — and found no gap and no over-acceptance, including the one MRI
rejects.

The section says in as many words that twenty-three hand-picked constructs are
not a completeness proof, because they are not.

Also

ast.go's Lines doc said "a position field on all 55 node types" and
the file defines 56. A count of something the file itself defines rots on
the next addition and tells the reader nothing, so it is gone rather than
corrected — I had reread that comment the day before, while adding Block.Line,
and did not check the number.

Green

go test ./... -count=1 exit 0 (4 packages), gofmt clean, and the parser
package stays at 100.0% of statements.

🤖 Generated with Claude Code